AI.info

The Pulse — Page 49

Browse The Pulse on AI.info.

• Waymo Closes $5 Billion Term Loan to Expand Robotaxi Service

  Waymo closed a $5 billion term loan, its first debt financing, to support expansion of its autonomous ride-hailing service in the United States and internationally. The announcement names 13 lenders and comes as the company prepares for launches in Tokyo and Singapore.

• AI Answers Are Becoming a Midterm Voter’s First Stop

  AI-generated answers and search summaries are already shaping how voters find election information, sometimes surfacing misleading pages or incomplete answers. Companies are adding official voting resources, but studies and reporting show that gaps remain.

• Lancet Commission Names Malicious AI an Extinction Risk

  A Lancet Commission identifies malicious use of artificial intelligence and nuclear conflict as threats with potential to cause human extinction by 2100. Its assessment says action to reduce malicious-AI risk could avert an estimated 10.4 billion disability-adjusted life-years.

• Nous Research Post-Trains Qwen3.8-27B and Calls the Model Arete-27B

  Nous Research says it used supervised fine-tuning to post-train Alibaba’s Qwen3.8-27B after observing refusals, deflections and CCP-framed answers on politically sensitive topics. In Nous Research’s 803-prompt evaluation, Arete-27B scored 2.98 ± 0.03 out of 5 on fact recall, while 0.9% of answers were classified as containing any propaganda.

• NVIDIA Opens Nemotron Models for Always-On AI Agents

  NVIDIA released Nemotron 3.5 Super VL and a 24GB version of Nemotron 3.5 Bolt on October 15, 2026, targeting enterprise and local AI agents. A smaller 4GB Bolt model is scheduled for October 23, while Super VL is available through Hugging Face, OpenRouter, ModelScope and NVIDIA NIM.

• Xiaomi Open-Sources MiMo-V2.6 Pro and Flash With Reinforcement-Learning Resources

  Xiaomi has released the MiMo-V2.6-Pro and MiMo-V2.6-Flash models alongside their weights, technical report and reinforcement-learning resources.

• Nuance Labs Raises $50M for Full-Duplex AI Avatars

  Nuance Labs has raised $50 million in Series A funding to develop a full-duplex audiovisual model for face-to-face AI conversations. The Seattle startup says its system can perceive speech, tone, gaze and gesture while generating vocal and facial responses in real time.

• TypeSafe AI’s Jev Reached Nearly 13% of Paid Teams Within 24 Hours, Vercel Says

  TypeSafe AI’s Jev reached nearly 13% of paid teams within 24 hours, more than twice the reach of any previous model launch, according to Vercel.

• Apple Tightens App Rules for AI Nudification and Bait-and-Switch Apps

  Apple updated its App Review Guidelines on October 16 to explicitly require apps to prevent and remove nonconsensual intimate imagery, including AI-generated imagery. A new rule also bars hidden or dormant features, with violations potentially leading to App Store removal and developer-account termination.

• India Orders Caller-ID Apps to Share Spam Reports

  India’s telecom regulator now requires caller-ID and call-management apps to send user spam reports to telecom operators. The rules also classify undeclared automated and AI-generated calls as spam and allow operators to charge for some application-to-person calls.

• TII Releases Falcon-Emirati-7B for Emirati Arabic

  The Technology Innovation Institute has released Falcon-Emirati-7B, a language model adapted from Falcon-H1-Arabic to understand and generate Emirati Arabic. TII says it scored 84.83% on its Alyah benchmark and is available through Falcon Chat.

• Anthropic Replaces Git With Scion After Merges Top 30,000 a Day

  Daily changes merged to Anthropic’s monorepo rose from about 2,500 in May 2026 to more than 30,000 in September. The company built Scion, an in-house version-control system, and moved the repository off Git in five months.

• Spain’s AEPD Logs First AI-Agent Data-Breach Report

  Spain’s data protection authority has received its first notification of a personal-data breach allegedly executed by an AI agent. The agency says the system found vulnerabilities, accessed an application, altered personal data and reached invoices, while warning that the incident still requires analysis.

• Personal AI Agents Steer Wealthier Users Toward Costlier Choices

  A new study reports that eight of 13 personal AI models recommended more expensive flights, insurance plans and graduate programs to wealthier users making identical requests. The effect persisted when users asked for the cheapest option and when agents inferred wealth from unrelated email.

• Google Gives Teachers AI Tools to Build Interactive Lessons

  Google Research has released more than 30 AI-generated learning interactives for middle and high school STEM subjects. The system lets teachers request custom simulations with guided challenges, hints and teacher approval built into the workflow.

• METR Estimates Anthropic’s Model Capabilities Are Advancing at About Twice Their Earlier Rate

  METR estimates that Anthropic’s models are improving on well-specified tasks at about twice their earlier rate, while warning that the estimate is uncertain. Its assessment also describes faster growth in benchmark scores and evidence that AI use may be contributing to the acceleration.

• Crusoe Plans 106-MW AI Cloud Campus Near Itaipu

  Crusoe and Paraguay’s Araico plan an AI cloud deployment near the Itaipu substation, with a path to 106 MW of IT capacity and possible expansion to about 280 MW. The companies target an initial launch in mid-2027, while questions about the project’s electricity arrangements remain unanswered publicly by Paraguay’s utility.

• Algolia Buys Velou to Enrich Product Data for AI Shopping

  Algolia acquired New York-based Velou on October 6, 2026, adding multimodal tools that turn product text and images into structured retail data. Velou’s enrichment will support search, recommendations, personalization and Agent Studio without requiring customers to rebuild existing integrations, the company says.

• Mistral Opens Munich Hub for Physics and Industrial AI

  Mistral opened a Munich hub on September 28, 2026, for research in Physics AI and industrial AI. The company says its teams will work with BMW, Siemens Energy and Technical University of Munich on engineering applications.

• Richard Borcherds Stops Grading Math Homework for Correctness

  UC Berkeley mathematics professor Richard Borcherds now gives homework credit for completion rather than correctness, saying AI use has made grading answers for accuracy seem pointless. He also told The Daily Californian he is uncertain what to advise students considering mathematics as a field.

• Single-Agent Loops Can Beat Subagent Systems, Anthropic Says

  Some enterprise agents perform better when specialist subagents move into a single agent loop with skills and tools, Anthropic says. Its new migration kit asks teams to compare each change against task-completion evaluations and keep subagents when they still help.

• More Than 300 Boston Dynamics Workers Challenge ICE Robot-Dog Deal

  More than 300 Boston Dynamics employees have urged executives to stop a deal supplying ICE with four modified Spot robots. The company says it will proceed and points to restrictions against using the machines to harm or intimidate people.

• MobileCybench Finds 23 Previously Unreported Android Vulnerabilities

  A benchmark using executable security probes tests how coding agents discover vulnerabilities in Android applications.

• Meta Adds Muse Connectors for Small Businesses

  Meta expanded its Muse AI agent with connectors for small-business software such as QuickBooks, Shopify and Slack. The agent can analyze business accounts and draft work, but Meta says owners must approve anything it publishes, sends or spends.

• Ai2’s FairStress Finds Bias Beneath Stable LLM Decisions

  Ai2 released FairStress, a benchmark that tests whether large language models change decisions unevenly when demographic cues and persuasive prompts are introduced. Its evaluation found that models could make similar initial choices yet respond differently to pressure, depending on which demographic group a prompt favored.

• NVIDIA Builds Robotaxi Push Around One Full-Stack Platform

  Robotaxi operators, automakers and autonomous-driving developers across Asia, Europe, the Middle East and North America are adopting NVIDIA’s DRIVE platform, the company says. The company’s September 10, 2026 update highlights Uber’s planned 28-city expansion, DRIVE Hyperion 10 hardware and the Alpamayo family of open driving models.

• Google Adds Agentic Video Understanding to Gemini

  Google DeepMind has added agentic video understanding to Gemini 3.7 Flash, 3.6 Flash and 3.5 Flash-Lite. The feature lets Gemini search video dynamically, cutting token use by up to 88% and analysis costs by up to 66%, according to Google.

• OpenAI Prices GPT-6.1 Sol at One-Fifth of GPT-6 Astra

  OpenAI says GPT-6.1 Sol approaches GPT-6 Astra on several agentic tasks at one-fifth of the standard input and output token prices. The company reports gains in coding, computer use and professional work, while saying Astra remains its recommendation for the most difficult scientific research tasks.

• UNIDIR Opens a New Forum on Military AI at UN Headquarters

  UNIDIR is launching a multistakeholder forum on artificial intelligence, peace and security at United Nations Headquarters in New York on October 22, 2026. The launch coincides with the opening of expressions of interest to join the forum, while separate Centre initiatives on industry guidance, AI terminology and non-state actors are also being introduced.

• AI Research Agents Match Just 18% of Human Ideas

  The IdeaTaste benchmark found that language-model agents matched human research ideas at 18% of the human level, compared with 26% for citations. Its authors tested more than 90,000 proposals across four research areas and found the largest gaps in concrete approaches and expected success.

• Researchers Extract Trained Neural Networks From Hard Labels

  A new study demonstrates end-to-end extraction of trained ReLU networks using only output labels from black-box queries. The method reaches at least 98.477% label agreement on MNIST and Fashion-MNIST test models.

• OpenAI’s Pando Backend Served 57TB of Git Data in an Hour

  OpenAI’s Pando Git backend served more than 57 terabytes of Git packs in one hour and nearly a petabyte of data per day, the company says. The system now powers development in OpenAI’s monorepo, where it supports concurrent work by engineers and Codex agents while preserving Git workflows.

• EU Publishes First Harmonised AI Act Standard, EN 18286

  The European Commission has published EN 18286:2026 in the EU Official Journal, giving providers of high-risk AI systems a standards-based route to show conformity with covered quality-management requirements. The standard addresses Article 17 of the AI Act and can be layered onto existing quality-management systems.

• Ultra Raises $50 Million to Expand Its Warehouse Robot Service

  Brooklyn-based Ultra raised a $50 million Series A led by Framework Ventures, bringing its announced funding to $62 million. The warehouse robotics startup charges customers an installation fee and monthly support fees, and says its robots have packed more than 500,000 orders.

• CrowdStrike Adds an AI Agent Kill Switch to Falcon Guardian

  CrowdStrike has introduced AI Kill Switch in Falcon Guardian, where security teams can block AI model access through connected gateways. Coordinated containment across endpoints, cloud, identities, SaaS and browsers is planned as a later extension.

• FAA Plans $875 Million SMART AI Program for Air-Traffic Operations, TechCrunch Reports

  The FAA plans to launch SMART, a $875 million AI software program for air-traffic operations, TechCrunch reports.

• Google Opens Gemini 4 Argon to Cybersecurity Defenders

  Google announced Gemini 4 Argon on September 30, positioning the model for software engineering, legal and financial work, and cybersecurity defense. The company is first giving access to trusted cyber defenders and says broader access will begin with paid API customers and Google AI Ultra subscribers.

• McDonald’s AI Pricing Tool Advises Nearly 14,000 Restaurants

  McDonald’s pricing engine analyzes millions of daily transactions and recommends menu prices for nearly 14,000 restaurants. Reuters found the system has widened some price differences between nearby stores and reported tensions over franchisees’ use of its guidance.

• UK and US launch AI partnership after uncrewed torpedo test

  The UK and US are launching a defence partnership focused on artificial intelligence and autonomy. The announcement follows the first launch of a US torpedo from a British uncrewed undersea vehicle.

• Microsoft Says AI Designed a Vanadium-Free Flow Battery Candidate

  Microsoft’s Discovery Engine helped Yale researchers design and validate a novel organic molecule for long-duration redox-flow batteries, the company says. The work uses AI agents, laboratory measurements and stored negative results to refine battery chemistry across multiple design rounds.